Hdf5 loading

Hi,

This MR allows the usage of bob.io.base.load for the batching using hdf5 files in the Bob format. For that I introduced this function from_file_name_to_tensor.

The implementation is not super clean; I had to introduce the a keyword argument called extension in the batching functions (https://gitlab.idiap.ch/bob/bob.learn.tensorflow/blob/3dcbd6afec902432152c17142dbd55ac0ca13f41/bob/learn/tensorflow/dataset/image.py#L10); but so far the thing does the job nicelly.

Mind to review that one ?

bob/learn/tensorflow/estimators/Siamese.py

@@ -139,7 +139,7 @@ class Siamese(estimator.Estimator):
                                                               self.extra_checkpoint["scopes"])
     
                 # Compute Loss (for both TRAIN and EVAL modes)
-                self.loss = self.loss_op(prelogits_left, prelogits_left, labels)
+                self.loss = self.loss_op(prelogits_left, prelogits_right, labels)
                
                 # Configure the Training Op (for TRAIN mode)
                 global_step = tf.contrib.framework.get_or_create_global_step()